This proposal describes new methods for protein secondary structure prediction, tertiary folding to low resolution using reduced protein models, and structural refinement using atomic level potential functions and a continuum model for the solvation free energy. The methods will be applied to ab initio prediction of protein structure, threading, loop refinement in homology modeling, and protein structure determination from sparse NMR datasets. Specific methodological areas of focus include: (1) new scoring functions for secondary structure prediction; (2) development of a reduced potential function, and global optimization methods for tertiary folding; (3) construction of a polarizable protein potential function using ab initio quantum chemical data; (4) improvement of Poisson-Boltzmann and generalized Born continuum solvation models; (5) inclusion of NMR constraints in the tertiary folding methodology.